Secondary school students’ gaze paths when interpreting graphs: an analysis with machine learning

Description

Can students’ eye-tracking data be used to automatically identify their strategy and predict their answer? If so, automated online feedback could be provided during a graph interpretation task. Automated answer prediction can be done with machine learning software. The difficulty of machine learning is that this is a black-box process that does not reveal how it predicts. We therefore developed a grey-box method that predicts students’ answers based on students’ gaze data and gives insight in students’ problem solving strategies. We do this within the theoretical framework of gaze data as evidence of conceptual actions (Chumamenko, Shvarts, Budanov, 2014).
Fifty students (age 15–19) were placed in front of a Tobii XII-60 eye-tracker and asked questions that required interpreting graphs. Based on the gaze data, machine learning software predicted whether students answered correctly (84% accuracy) showing that it is possible to predict students’ answers based on their gazes and therefore worth constructing a new model that gives insight into students’ strategies. This grey-box model was informed by previous research (Boels et al., 2018). For this model we divided the gazes into vertical (correct strategy) and horizontal (incorrect strategy) eye movements. Our model predicts with the same accuracy as the black-box machine learning but with the benefit of providing insight into a key parameter that helps predict answer correctness.
The results of this research can be used in the development of online student feedback systems based on students’ gaze patterns. Our method is an addition to the toolbox of eye-tracking research.

Period	16 Dec 2020
Event title	EARLI SIG 27 conference on online measures
Event type	Conference
Location	Antwerp, BelgiumShow on map
Degree of Recognition	International